Probabilistic electricity demand forecasting for Türkiye's 2035 energy plan: A risk-based framework for policy and capacity planning

In emerging economies marked by structural transformation and macroeconomic volatility, inaccurate electricity demand forecasting poses dual threats: supply-security risks from underinvestment and fiscal inefficiencies from overinvestment. For Türkiye, where energy planning centers on the 2035 National Energy Plan (UEP-2035), quantifying these risks is essential. This study develops a policy-oriented probabilistic forecasting framework for Türkiye's electricity demand to 2035. It integrates Lasso, Ridge, SVR, and ANN with Monte Carlo simulation (10,000 iterations) on a multivariate dataset (1990–2023). The dataset combines macroeconomic indicators with satellite-derived Nighttime Light (NTL) as a proxy for economic activity. SVR was chosen as the primary model due to its superior cross-validated performance (CV RMSE = 0.029746) and strong predictive-interval coverage on withheld 2019–2023 data. The framework generated probabilistic demand distributions (P10, P50, P90) for 2024–2035. For 2035, P50 estimates range from 485.43 to 518.83 TWh, with the UEP-2035 target of 510.5 TWh between the median and 90th percentile. P90 scenarios show exceedances of 10.0–17.9% above the target, implying 11–22 GW of additional capacity under empirical capacity factors. A non-parametric block-bootstrap test raises the ensemble P50 to 536.5 TWh. It widens the band from 103.0 to 116.9 TWh, indicating central estimates are sensitive to shock-distribution assumptions while policy implications remain directionally informative. By linking probabilistic forecasts to policy benchmarks, this study shifts focus from prediction accuracy to decision robustness. The framework offers a transparent and replicable decision-support tool for energy policy in emerging economies facing data scarcity and macroeconomic uncertainty.